GitLab Atlas
100% GitLab Duo Agent Platform Native — Zero Config, Zero API Keys
Your repo has code. Atlas gives it a voice — with 6 artifacts, Mermaid + PlantUML diagrams, and drift governance.
Works with any GitLab repository. No API keys needed — powered by GitLab Duo natively.
The Difference
See What Atlas Adds to Your Repo
One mention generates 6 architecture artifacts with Mermaid and PlantUML diagrams.
Undocumented Repository
Atlas-Powered Repository
Watch It Happen
See Atlas Get Enabled Right Inside GitLab
Instead of reading menu instructions, watch the real path: open AI Catalog, switch to Flows, and select Atlas for your project.
vibeVoice
Voice-powered AI companion — Atlas demo target project
Atlas Refresh
Refreshes repository architecture artifacts after merge or pipeline events
Atlas Navigator
Explains MR changes and detects architecture drift with visual diagrams
Atlas Bootstrap
Generates living architecture baseline for any repository
Atlas Bootstrap
Generates living architecture baseline for any repository
Visibility & access
Managed by GitLab AI Hackathon / Participants / sejun1
Service account
Configuration
1 version: v1 2 environment: ambient 3 components: 4 - name: analyzer 5 type: AgentComponent 6 - name: generator 7 toolset: create_file_with_contents 8 - name: validator
Show the actual GitLab flow path, not a generic catalog mock.
This mirrors your real screenshots: project home, Automate, Flows, AI Catalog search, Atlas results, and the Bootstrap detail page with Enable visible.
Quick Start
Up and Running in 3 Steps
No code changes. No API keys. Published to AI Catalog — any project can enable.
Enable from AI Catalog
Find Atlas flows in the GitLab AI Catalog and enable them for your project. No fork needed.
AI Catalog → Atlas → Enable
Mention in Issue or MR
Trigger any Atlas flow by mentioning it in an issue or merge request comment.
@ai-atlas-bootstrap-{group}
Review the MR
Atlas creates a feature branch, generates 6 artifacts, and opens a merge request for your review.
Branch + MR → Review → Merge
Setup Guide
Enable Atlas on Your Project
Follow these steps in the GitLab UI to enable Atlas flows on any project.
Enable Duo Agent Platform
Group-Level · One-Time- Navigate to your group's Settings → GitLab Duo
- Select Change configuration
- Toggle "Turn on GitLab Duo Chat (agentic), agents, and flows"
Enable Flows from AI Catalog
Project-Level- Go to your project → Automate → Flows in the sidebar
- Select Enable flow from group (upper-right button)
- Choose "Atlas Bootstrap" and set triggers:
Mention+Assign - Repeat for "Atlas Navigator" and "Atlas Refresh"
Configure Runners
If Self-Managed- Flows require runners with the
gitlab--duotag - On GitLab.com SaaS: hosted runners work automatically
- Self-managed: assign runners with
dockerexecutor to the top-level group
Trigger Atlas
Ready to Go- In an Issue:
@ai-atlas-bootstrap-{group} bootstrap - In an MR:
@ai-atlas-navigator-{group} explain - Results appear as comments with rendered diagrams
{group} with your top-level GitLab group name.Capabilities
Three Agent Flows, One Architecture
Each flow is a GitLab Duo Agent Platform workflow published to the AI Catalog.
Atlas Bootstrap
Analyzer → Generator → Validator (3-component flow). Scans your repo, reads key files, generates 6 architecture artifacts (TECH_STACK, ARCHITECTURE with ambiguities & improvements, DATA_FLOW Mermaid, 2 PlantUML diagrams, README section), and creates a merge request.
Atlas Navigator
Explains MR changes with before/after diagrams, detects drift with governance decisions (allow / warn / require_docs_update / require_follow_up), and auto-creates follow-up issues for high-severity drift.
Atlas Refresh
Automatically updates architecture documentation after merge to keep the baseline current. Creates a new MR with refreshed artifacts so your docs are never stale.
Web Dashboard
GitLab OAuth login, project list with one-click Bootstrap, Architecture Artifact Browser (renders Mermaid + Markdown in browser), and Atlas Status Panel with 4 health checks (flows, service account, docs, CI).
Live Demo
Real Output from a Real Project
Atlas ran on vibeVoice — here is what it produced.
Bootstrap Output
6 architecture artifacts generated
- MD TECH_STACK.md Languages, frameworks, deps
- MD ARCHITECTURE.md Design + ambiguities + improvements
- MMD DATA_FLOW.mmd Data flow diagram
- PUML DEPLOYMENT_BOARD.puml Infrastructure topology
- PUML SEQUENCE_FLOW.puml Request sequence diagram
- MD README.md Architecture section injected
DEPLOYMENT_BOARD.puml — rendered from actual Atlas output on vibeVoice
Navigator Analysis
MR comment posted by Atlas Navigator
System Design
Duo Agent Platform Architecture
Pattern B — create_file_with_contents + run_git_command. Feature branches and MRs, never direct main commits.
AI Catalog (Public)
Published flows — any GitLab project can discover and enable Atlas agent flows
Duo Agent Platform Flows
Bootstrap / Navigator / Refresh — triggered via @mention in issues or MRs
Analyzer → Generator → Validator
Claude Sonnet 4.5 Vertex — 3-component flow: analyze codebase, generate artifacts, validate output
Dual Diagram Engine
Mermaid (data flow) + PlantUML (deployment, sequence) — rendered inline in MR descriptions and GitLab UI
Web Dashboard
FastAPI + Jinja2 on Cloud Run — Architecture browser, status panel, OAuth login
By The Numbers
Atlas at a Glance
Any GitLab project can enable Atlas flows from the AI Catalog and generate living architecture docs.
Built With
Technology Stack
100% GitLab-native. Default LLM managed by GitLab — no external APIs, no configuration needed.
FAQ
Frequently Asked Questions
Common questions about Atlas setup, compatibility, and usage.
Which GitLab plans support Atlas?
Does Atlas work on self-managed GitLab instances?
gitlab--duo tag assigned to your top-level group with a docker executor. On GitLab.com SaaS, hosted runners are pre-configured and work automatically.Which programming languages are supported?
Will Atlas overwrite my existing documentation?
docs/atlas/ directory and opens a merge request for your review. Your existing files are never modified without explicit approval through the MR process.
Architecture Change Analysis
This MR introduces significant architectural changes across multiple system layers.